The R Project in Statistical Computing

R is an Open Source environment for statistical computing and graphics

‘not unlike’ version 3 of the S system developed at Bell Laboratories
(formerly AT&T, now Lucent), which is the basis of the commercial system
Brian D Ripley
S-PLUS® 1 from the Insightful Corporation.2 This article describes the R
Department of Statistics, philosophy and current status: it is definitely not a review as the author is a
University of Oxford member of the R core team as well as the S-PLUS Advisory Board.

ripley@[Link] What is R?

Executive summary: R is an advanced statistical computing system with very

high quality graphics that is freely available for most computing platforms.
More details are available on-line at [Link]

R currently exists as source code, as well as binary distributions for most

distributions of Linux, for Windows (95 and later: see Figure 1) and for the
PowerPC Macintosh. It can be downloaded from CRAN, with a master site
at [Link] and a series of mirrors including one at Bristol,
[Link]. It is distributed under the GNU Public Licence,
which means that it is freely available, both in the sense that it is available
for only the distribution costs and that the source code is open to view and
modification.

Fig.1: Part of an R session on Windows

The R project was started by Ross Ihaka and Robert Gentlemen at the
University of Auckland, for teaching on Macintoshes which by today’s
standards were very limited. R still runs well on low-end systems: a basic
installation takes around 20Mb of disc space and runs in 16Mb RAM. A usable
Windows distribution fits on three floppy discs, although these days people are
more likely to distribute CD-Rs or ask students to download it. The usefulness

1
Versions up to S-PLUS 2000 are based on S version 3; the current Unix/Linux version
6.0 of S-PLUS and the upcoming version 6.0 for Windows are based on S version 4.
2
Until January 2001 known as MathSoft Inc.

23
 The R Project in Statistical Computing Brian D Ripley

of the system is greatly enhanced by extension packages, smoothing, multivariate analysis, neural networks, time
eight of which ship with R and about 100 are available series... Classical statistics is also covered; one of the
for download from CRAN. This download can be done extension packages is devoted to classical tests, for
easily from within R: for example example.

[Link]("VR", There are now many examples of R being used for

CRAN="[Link]") serious large-scale data analysis. It has been used for
election forecasting in Austria and will be used in the
will download (if you are on-line) and install the R UK; it is used for a number of systems in the processing
support software for Venables & Ripley (1999). of gene expression data, and my group have used it to
R provides a command-line-driven interpreter for the S process time series of brain images.
language. The dialect supported is pretty close to that
implemented in S-PLUS 2000, close enough that many There are projects providing a Web-based interface to R.
users will not notice the differences. They will notice Rcgi ([Link] by
some differences in output: the R developers have M. J. Ray at the University of East Anglia provides a form
different preferences. interface for students to submit data or code to a copy of
R running on the web server and get back numeric and
What is it being used for? graphical output.

With a freely-distributed system it can be hard to know R sounds a lot like S-PLUS!
what it is being used for and how extensively. One clue
comes from requests for support. A free system such as To make a comparison more precise, let us confine
R is supported by an email list, and that suggests that R attention to S-PLUS 2000, the current Windows version.
is used extensively in elementary teaching, for student R and S-PLUS 2000 are similar, and will be used in very
projects and surprisingly often by researchers including similar ways in the teaching context. Their languages
those in companies. are similar enough that students who use S-PLUS in the
university can practise with R at home (and it is clear
At present R is almost entirely driven from a command from the email support list that many do). A few
line, even on versions with GUI consoles (Windows, comparisons:
Mac, GNOME under Linux). It is hotly debated whether a) S-PLUS 2000 is commercial and comes with
this is a disadvantage in teaching, as students are commercial support. R is much easier to distribute
increasingly used to filling in a series of point-and-click b) R is much smaller and runs on less powerful
forms with buttons to click for help. R is not like that, machines
and although there are ways to customize it with such c) S-PLUS 2000 is monolithic: R has a very small
interfaces (by programming in C, Tcl/Tk or Java), these core and many extensions
are laborious to use. My own department decided not d) S-PLUS 2000 does have a ‘menus and dialog
to use it for undergraduate classes because a GUI boxes’ interface as well as a command-line
interface was thought easier to support. The obverse interface. It also has ways to edit graphs
side is that a command-line interface is a much richer e) There are conflicting claims about performance. I
environment, and students are forced to think about find them about equal, but R is more tolerant of
what they are doing when they use it. That debate will badly-written code which can make S-PLUS slow
run and run. to a crawl
f) There is not much to choose in quality. S-PLUS is a
Nolan & Speed (2000) show how R can be used in more mature system. I suspect that R has more
teaching an introductory statistics course. They integrate bugs, but that these will get fixed much faster and
the theory of statistics with the practice of statistics by subject-matter experts
through a collection of case studies, and use R (or S- g) Both have 2D graphics of much higher quality
PLUS) to analyse the data. This is an inspiring book that than most commercial statistical systems. R is
I highly recommend to those planning to teach real particularly good at annotating with mathematical
statistics. notation (see Figure 2). S-PLUS has much richer
facilities for multi-panel plots (Trellis™)
R is an advanced system: it provides statistical methods h) Both are weak on ‘graphics for the video games
up to [Link]. level and beyond. Many research students generation’, that is 3D and dynamic graphics, but
use it as a research tool. It provides access to state-of- R is weaker
the-art methods in robust methods, density estimation,

24
 The R Project in Statistical Computing Brian D Ripley

computing, and in particular as one component in a

distributed computing system.

Another possibility is that commercial users will want

new features enough to pay for them to be developed;
some aspects of graphics have been mentioned.
Using R as a computational engine in other systems is
relatively easy and for open systems has no licence
complications. There are ways to embed R into a, say,
Visual Basic front-end and to call R functionality from
Excel. This provides a rich vein to be explored, and
Fig.2: Mathematical annotation
now has Unix/Linux analogues, too.
This figure shows the expected value of the residual
deviance for a sample from Poisson(l), and hence that Much of what is now available to all in R comes from
apparent under- or over-dispersion can often be attributable
the teaching needs of a few. Contributions will be
to bias.
welcomed with open arms.

References
Why Open Source?
[1] Nolan, D. & Speed, T. P. (2000) Stat Labs:
R may be free to the end user, but its development has Mathematical Statistics Through Applications,
involved man-years of volunteer effort by the core team Springer. Support material is available at:
(about a dozen people, not all active at all times) and [Link]
contributors. We are sometimes asked why we do it. One [2] Venables, W. N. & Ripley, B. D. (1999) Modern
motivation for me has been to make available a first-class Applied Statistics with S-PLUS. Springer. Support
statistical system to those with fewer resources, in particular material is available at http://
students and researchers in the third world. Now when [Link]/pub/MASS3/, including
students from developing countries ask Bill Venables or Complements describing its use with R.
me for a copy of S-PLUS, as they often do, we can point
them towards R in the knowledge that they will be getting
a very useful product for an unbeatable price. (When they
ask for free, open source versions of our books as well, we
can only shrug.)

An advantage for me is to be able to study all the source Applied regression including
code, so I can find out (and by now probably have) computing and graphics
precisely which variation on a statistical method has been
implemented. There used to be an adage that to really
Sanford Weisberg, University of
learn a method you had to know how to do it by hand (and
Minnesota
students were made to do so). The current analogue is to
understand what the source code does.
Workshop, 3 May 2001, at the
Where is R going? University of Glasgow

R has been developing very rapidly. Until the first non- Further details from Maths, Stats & OR
beta release, version 1.0.0 on 29 Feb 2000, there was Network in Glasgow
about a release a month. The aim is now for a full release
about twice a year with intermediate bug-fix releases.
We are still adding statistical features (most recently
tel 0141 330 2718
MANOVA and factor analysis at the request of psychologist
users), and we have been addressing performance issues. email anne@[Link]

As a volunteer project, where R goes depends entirely on [Link]

what its volunteers want to do. My impression is that it
will now move fastest as a research tool in statistical

25